Before reading, check out this wonderful old flash game from which I blatantly stole this concept: http://www.ferryhalim.com/orisinal/g2/dozen.htm (Also that site has some ridiculously adorable stuff. Check it all out. But first read my post and help me ok good thanks cool.)

The concept of the game is simple: you're in a basket, you have to jump straight up to land in the next basket. Jumps are always perfectly vertical an always the same height. In that game, the layout is all predefined and it has an end. I started thinking about how one might randomly generate the next basket to jump to in real time in order to make an infinitely scrolling game. This seemed simple at first, and certainly could be simple with an assortment of predefined cookie-cutter paths for the baskets to follow that could be used, but that doesn't make things interesting. When considering generating arbitrary paths or paths arbitrarily modified from a set of "template" path shapes (rather than exact predefined paths) with varying basket speeds and basket starting positions, the problem becomes a lot more complicated.

The first goal is to be able to determine whether a given pair of baskets makes a valid level, and then to assign the level a "difficulty" score. To do this we determine the periods of the two basket paths given path length and basket speed. Then we search through time from 0 to the LCM of the two periods, looking at the X and Y of the paths in order to locate all contiguous ranges ("windows") of t values at which jumping would land you in the upper basket. Given the set of jumping windows, difficulty can be derived with a function of number of windows total, size of windows, and time between windows. The actual difficulty scoring function isn't important here, but that's easy enough to create once we have the window data.

Let's consider a simplified case using only one positional dimension, y. Imagine a stationary basket on path y=1 directly below a basket moving up and down along path y = sin(2t) + maxJumpHeight. Half the time, the basket will be within jumping range (when y_upper < y_lower + maxJumpHeight) and the other half it will be too high. The LCM of the period of the paths is ~3.14 timesteps (seconds? sure), considering stationary paths as having a period of 1 not 0. Now for every time t in range [0,3.14] we can plot the y values of the two baskets at t. Now let's say it takes ~0.7s for the ball to reach a maxJumpHeight of 2 or so. Now we can easily plot the peak jump height for a jump starting at any given t value. This can be plotted as y = y_lower(t-0.7) + 2

Plotting those should look something like this:

Blue = upper path over time
Orange = lower path over time (slanted for visibility because colors are dumb)
Red = ball max height over time (slanted to show how offset lines up)
Any point at time ton the red line actually represents the peak of a jump if you had jumped at time t-0.7. As long as we know this, we can ignore that offset as we only care about the ball's path and the upper basket's path at time t.

This implies that there are 2 places where you can land in the upper basket: intersections at points A and B. But that assumes you can ONLY land at the peak of your jump. We actually need to add another dimension to this because we care about all points along the ball's falling trajectory, not just the peak. To add this additional dimension, we must consider how long the ball has been falling for: a jump offset axis j. This axis projects the falling motion forward in time for each second of j since the peak of a jump. Projecting this through the sine wave path of y_uper extended into the j dimension, we can see that there are many collisions along the falling path.

Because we care only about points after the jump peak, instead of projecting the jump paths forward in time for each timestep of j, I decided to project the falling arcs straight out along the j axis and project the basket path backwards for each timestep of j. These should be equivalent transformations. By projecting the falling arcs straight along the j axis,

I know this is confusing as heck. I've tried to make these graphs as clear as possible but it's hard with what I'm using...
X axis = t; Y axis = y; Z axis = j

(I messed that graph up 3 times before even posting this, so I hope this time it's right)

Note that the sine wave position projects backward through X (t) for each Z (j) timestep, and the parabolic arcs project straight in the Z (j) plane. This should be equivalent to the sine wave projecting straight out in the Z axis and the jump arcs projecting forward through X for each Z step, right? I think that's right.

Hopefully you can tell that at SOME point along the right hand side of the graph there is an intersection between the half parabola of the falling arc and the position vector of the upper basket. Any intersection implies that jumping at time t-0.7 would land you in the basket at time t.

THEREFORE, a valid value of t is one at which there is an intersection at any j value after the peak of the jump.
In this graph, the "jump window" would be the second half of the interval. You could jump any time during that window and you would still land in the basket.


Ok, now that we've done it with a 1D position axis, add another one. Instead of only considering the Y value, we need to consider X and Y for the jump arcs and the basket paths at each time and jump offset. This makes the resulting graph 4-dimensional, meaning 3D intersections at any given t value rather than 2D.....yay....

I feel like I simply can't wrap my head around how to do any calculation for this (even in the simplified case) to effectively locate t values that contain intersections without just iterating over each one and running a 2D/3D intersection on the functions using j as the sole independent. Is there some sort of magic I could be using here?

---

I should also mention that this doesn't even begin to cover GENERATING the next path, just locating the jump windows for a given pair of paths...It's possible that it could be easier to generate a path that is guaranteed valid rather than validating an existing path, though I'm not sure how to figure that out either.


Thanks for taking the time to read all this. If there's any way I can clarify things, please ask. I think I have all the pieces in my brain to set up the situation, I just don't have the math to solve it, so hopefully I can explain any questions you might have.

-Taw

Heh well I guess the first step is triangulation, THEN opaque triangles.

Do you have any ideas on how you're going to specify triangles? I feel like the user would manually select a triangle by choosing the three vertices, and THAT's when the model should be gray. Points and edges would turn black once selected. Maybe you can use like a gray-black flashing cursor or something for that drawing so that it contrasts with both the gray and the black models.

The fact is that you'll almost definitely want user-selected triangles, rather than having to write any triangulation routines for arbitrary polygon selections (which would be a total bitch to write). You COULD write some triangulation routines, but I wouldn't recommend it.

graying out the cursor is done in the grayscale version, so do you want a combination of the two where the vertices stay black but the cursor is gray?

When I add line drawing or selecting multiple vertices, would you prefer the points to stay black and the selection being indicated by multiple gray cursors? Or add gray cursors to the selected points, maybe with the real cursor being black, or do you prefer to have it all in b&w?

Yes, I'd say vertices stay black but cursor is gray, and then in multiple vertex selection I'd say either multiple gray cursors, or do the opposite and have one of them be gray point on a black cursor. It's just the contrast so that you can see the actual position of the point without the cursor getting in the way. It's totally functional the other way as well, it's mostly an issue with the further points where the length of the cursor lines are shorter.

It definitely looks great there in that Arwing video. Having everything in the model be black makes it a lot easier to see what you're doing and what you're working with.

I guess the next step is opaque polygons :O

That looks great! I think you should leave the sprites that are visibly adjacent to a wall as your originals and use the freestanding spikes as Keoni's. Also I noticed that you changed the location of that gap so it didn't have the spike issue. looks awesome!

Just a point I feel I should make after watching that screenie, one thing that VVVVVV makes a point to do very well is to never place spikes in a location that immediately kills you without warning when changing screens. There should never be a place where the spikes go all the way to the edge of the screen where you can enter from the other side. There should be at least a 1 tile gap so it doesn't feel like the game is cheating you.

Other than that this game honestly looks fantastic!

The 01100110011 pattern is the value assigned to the bit in b_nodes.

You mark the starting location as closed. Then EVERYTHING that is closed so far is expanded into the surrounding nodes that aren't closed yet. Each of those are then marked as closed and given the current pattern value.

Each pass over the buffer the value is either a 1 or a 0. Every bit expanded during that pass is assigned the same value. The value then changes for the next pass to the next value in the pattern.

This doesn't work for diagonal movement because of the pattern. That pattern only works for 4-directional movement. There may be multiple paths found to the goal, but all of them will have the same length, so it doesn't matter which one you choose.

I'll make some pictures to illustrate the concept.


PICTURES:

Here you can see that it expands every possible node each pass. The value assigned follows the pattern I described. Once the goal is found, you can follow the pattern in reverse to find your way back.

That was the path I chose to follow, but any of these paths would be valid, because they all follow the pattern:

Note that they are all the exact same length. So technically the set of valid tiles to move to is:


EDIT: I put the pictures in the first post as well

You definitely do not need advanced pathfinding of any kind for this.

Something thing to note is that most enemies who move deliberately closer to you don't begin to do so until you're within a certain distance from them, within their line of sight, etc. Otherwise, they just move randomly as well. Even so, the best they can do is simply move toward you, they never actually do any pathfinding. Ever. EVER. (Ever.)

Also........Which Zelda game specifically is this a clone of? Is it Link's Awakening? Or an older one than that?

What are you using for sprites in this? Are those original? Or modified versions of the sprites used in the original TinyCraft thread?